fix: preserve German umlauts for proper transliteration

[abhu85]

Summary

Fixes #50 - German umlauts (ä, ö, ü, Ä, Ö, Ü, ß) are now correctly transliterated to their digraph equivalents (ä→ae, ö→oe, ü→ue, ß→ss).

Before: München → munchen
After: München → muenchen

Problem

The NFKD normalization step was stripping German umlauts before speakingurl could apply its proper character mappings. For example, ü was being normalized to u (removing the combining diaeresis), so speakingurl never saw the original character and couldn't map it to ue.

Solution

Preserve German umlauts (and any characters specified in custom mappings) during the normalization step using placeholders, then restore them before passing to speakingurl. This allows speakingurl's comprehensive charMap to handle German characters correctly.

The normalization is still applied to other characters (like macrons) where speakingurl's handling is less ideal.

Test plan

• Added new test suite for German umlaut handling
• All 11 tests pass
• Verified no regression in macron handling (Pōneke → poneke)
• Custom mappings still override default behavior (München with {custom: {'ü': 'u'}} → munchen)

[lovell]

This looks like it was LLM-generated and overly verbose.

Could the fix be as simple as not removing the combining diaeresis?

-  const deburredText = text.normalize('NFKD').replace(/[\u0300-\u036f]/g, '');
+  const deburredText = text.normalize('NFKD').replace(/[\u0300-\u0307\u0309-\u036f]/g, '');

[abhu85]

Thanks for the feedback @lovell! You're right - the PR was AI-generated and the original approach was overly verbose.

I've updated the PR with your simpler fix. I also added .normalize('NFC') after the selective removal to recompose the characters back into precomposed form (e.g., o + U+0308 → ö), since speakingurl expects precomposed characters for its mapping.

const deburredText = text.normalize('NFKD').replace(/[\u0300-\u0307\u0309-\u036f]/g, '').normalize('NFC');

Verified it works:

Schöner Städte Übung → schoener-staedte-uebung
München → muenchen  
Größe → groesse
Ärger → aerger

Thanks again for the better approach!

[lovell]

Thanks for the update, are you able to add a couple of basic test cases to help prevent future regression?

[abhu85]

Added test cases in commit 068299b:

ava('German umlaut transliteration', function (t) {
  t.plan(4);
  t.true(limax('Schöner Städte Übung', { lang: 'de' }) === 'schoener-staedte-uebung');
  t.true(limax('München', { lang: 'de' }) === 'muenchen');
  t.true(limax('Größe', { lang: 'de' }) === 'groesse');
  t.true(limax('Ärger', { lang: 'de' }) === 'aerger');
});

All 11 tests pass.

[lovell]

Thanks, does this work without the { lang: 'de' } option? A test case for that would be good also.

[abhu85]

Yes, it works without { lang: 'de' } as well. Added 4 more test cases in commit e7e2e85:

// Without lang option (default behavior)
t.true(limax('München') === 'muenchen');
t.true(limax('Größe') === 'groesse');
t.true(limax('Ärger') === 'aerger');
t.true(limax('Schöner Städte Übung') === 'schoener-staedte-uebung');

All 11 tests pass (8 German umlaut tests total - 4 with lang: 'de', 4 without).

[lovell]

Thank you / danke